Audio Uploads in ChatGPT Explained: Transcribe a Meeting, but Not on the Free Plan

If you have ever finished a meeting with a recording and no notes, this is the feature you were waiting for. On October 6, 2026 OpenAI's ChatGPT release notes added audio uploads: attach a recording to a chat, and ChatGPT will transcribe it, summarise it, or answer questions about what was said in it.

There is one catch, and OpenAI states it without hedging. Audio uploads are not available on the Free plan. Everything else about them, including which formats work and how big a file can be, is published on OpenAI's own help page, so this page is the whole answer rather than a guess.

Key Takeaways

• You can upload a recording and get a transcript back, as of October 6, 2026. OpenAI's wording: "Upload audio files to ChatGPT to create transcripts, summarize recordings, and ask questions about their contents."
• The free plan is excluded, in OpenAI's own words. Its help article says "Audio uploads are not available on the Free plan at this time." Free accounts keep document, spreadsheet, presentation, text and image uploads.
• Every paid subscription is covered, and OpenAI names no individual tier. The sentence is "available with paid ChatGPT subscriptions and workspaces, including Enterprise", which makes Go at $8 a month the cheapest route in.
• Audio files can be up to 512 MB. That is the same per-file cap OpenAI applies to documents.
• There is no published maximum length. Long recordings "may be processed in smaller sections", processing is "best-effort", and "very long recordings may time out."
• Nine audio formats work, including MP3, M4A and WAV. Video-identified WebM and MP4 files do not, even if all you want is the sound.
• OpenAI warns about accuracy twice. Transcripts "may contain errors", speaker identification "may be unreliable", and performance "may vary across languages".

Which Plans Get Audio Uploads

Paid plans get audio uploads, the Free plan does not, and OpenAI draws the line in a single sentence rather than a plan matrix. Its help article says: "Audio uploads are available with paid ChatGPT subscriptions and workspaces, including Enterprise. Audio uploads are not available on the Free plan at this time."

Two honest qualifications about that table. The first is that OpenAI never writes "Go" in the audio section, so Go is included because it is a paid subscription, not because OpenAI listed it. The second is OpenAI's own caveat, which it repeats in both places: "Availability may vary by workspace settings, region, client version, and selected model." If you pay and the attachment is refused, that is the sentence to blame, and updating the app is the first thing to try.

Prices above are from OpenAI's pricing documentation, read live on October 11, 2026.

Which Audio Formats ChatGPT Accepts

Nine audio formats work, and the list covers every format a phone, a laptop or a voice recorder is likely to produce. OpenAI publishes it as: "WAV, MP3/MPEG, OGG/OGA, audio-only WebM, PCM, FLAC, AAC, M4A, audio-only MP4."

That last row is the one that will catch people out, because it looks like it should work. OpenAI spells it out twice, once in the format table and once in its troubleshooting section: "WebM and MP4 files identified as video are not supported by audio uploads." It also adds that "the file must contain a valid, decodable audio stream."

The practical consequence: if your recording is a video file, extract the audio first. Exporting the audio track as an M4A or MP3 is enough, and most meeting tools offer an audio-only download alongside the video one.

How Big and How Long a Recording Can Be

A file can be up to 512 MB, and OpenAI publishes no maximum duration at all. The exact wording is "Audio files can be up to 512 MB. Longer recordings may be processed in smaller sections when Data Analysis is available. Processing is best-effort, and very long recordings may time out."

That is not the same as a time limit, and it is worth being precise about the difference. 512 MB is a lot of speech: a one-hour meeting recorded as a 128 kbps MP3 is roughly 55 MB, so the cap is generous for anything you are likely to record on a phone. What OpenAI is warning about is the processing end, not the upload end. "Best-effort" and "may time out" mean a very long recording can fail without breaking any published rule.

If you have a three-hour recording, the sensible approach is to split it rather than find out. Nothing OpenAI publishes promises it will come back.

What You Can Actually Ask For

Four jobs, and OpenAI names all four itself. Its release-note entry and help article between them list a transcript, a summary, structured notes and a follow-up email draft, which is close to the complete list of things people want from a recording.

• A plain transcript. "Ask for a transcript of an audio recording, then ask follow-up questions about its contents."
• A meeting summary with the decisions pulled out. "Summarize a meeting recording, highlighting decisions, open questions, and next steps."
• An answer to one specific question without reading the whole thing. OpenAI's own example: "What did they say about the launch deadline?"
• Notes or a draft email. "Turn a recording into structured notes, a meeting recap, or a follow-up email draft", and from the release note, "turn a meeting, interview, or lecture into structured notes".

It also lists extracting "key details from an interview or lecture, such as requirements, examples, or main takeaways", which is the student use case in OpenAI's own words.

The Accuracy Warnings, Which Are OpenAI's Own

OpenAI warns about transcript quality three separate times, and none of the warnings come with a number. There is no published accuracy rate, no supported-language list and no word-error figure, so the honest answer to "how good is it?" is that OpenAI has not said.

What it has said, verbatim:

• "Audio understanding and transcription performance may vary across languages."
• "Transcripts may contain errors, and speaker identification may be unreliable. Review important details against the original recording."
• From the release note: "Transcripts may contain errors, and performance may vary across languages."

Speaker identification is the one to take seriously. A transcript that puts the right sentence next to the wrong person is worse than no transcript in a meeting with four people in it, and OpenAI is telling you in advance not to trust that part. Treat the "who said it" column as a draft and the "what was said" column as reliable but checkable.

Where the File Goes Afterwards

Into your Library, where it stays until you delete it, and deleting the chat does not delete it. Files you upload to ChatGPT are saved automatically, which matters more for a recording of a private conversation than it does for a PDF.

The controls are the ones that already applied to every other upload, and we cover them in full in ChatGPT Library, explained:

• To delete a recording, delete it in Library, not just the chat. Open Library, select the file, choose Delete or the trash icon. Deleted files are scheduled for permanent deletion within 30 days.
• To keep a recording out of Library entirely, use Temporary Chat. OpenAI says "files uploaded while using Temporary Chat are not saved to your account or Library." Our guide to ChatGPT's temporary chat controls covers what else that mode changes.
• To keep a recording out of model training, turn off "Improve the model for everyone" in Settings, then Data Controls. OpenAI says it may use uploaded content to improve model performance when that setting is on.

Should You Pay For This?

If you record meetings, interviews or lectures regularly, Go at $8 a month is the cheapest plan that unlocks it, and it is the clearest single reason to leave the free tier that OpenAI has shipped in months. One meeting a week turned into notes you did not type is worth $8 to most people.

If you record something once a quarter, do not upgrade for this. Free dictation and note apps already on your phone will handle an occasional memo, and ChatGPT's advantage here is the follow-up questions, not the transcription itself.

If you were going to upload a recording of other people talking, check the room first. The file is saved to your account by default, the transcript is best-effort, and OpenAI itself says speaker labels may be wrong. Temporary Chat is the right place for anything sensitive.

If you are on Free and this is the feature you want, know that OpenAI said "at this time". That phrasing has preceded a wider rollout before, and if it changes we will update this page. Whether the $20 Plus plan earns its price over Go is a separate question, and we weigh it in is ChatGPT Plus worth it in 2026.

For the other half of ChatGPT's audio story, the model that handles your voice going in rather than a file, see ChatGPT's dictation model, explained.

Sources

• OpenAI Help Center, ChatGPT release notes, entry dated October 6, 2026, "Audio uploads in ChatGPT", for the feature description, the four use cases and the paid-plan availability line. Read from a capture dated October 10, 2026.
• OpenAI Help Center, Uploading files and audio to ChatGPT, for the Free-plan exclusion, the nine audio formats, the video-file exclusion, the 512 MB cap, the processing caveats, the 80-files-per-3-hours and 3-per-day rates, and the accuracy warnings. Read from a capture dated October 7, 2026, carrying "Updated: 21 hours ago".
• OpenAI Help Center, Using Library to manage files in ChatGPT, for file retention, deletion and the Temporary Chat exclusion. Read from a capture dated October 8, 2026.
• OpenAI, ChatGPT pricing documentation, for the $0, $8, $20, $100 and $20-per-user plan prices. Read live October 11, 2026.